Picture this: your mom proudly tells you she finally bought ChatGPT. You take a look, and it’s not ChatGPT at all. It’s a copycat app called “Ask AI” with a logo designed to look just confusing enough. She just paid $40 for nothing.
That actually happened, and it kicked off one of the more useful livestreams I’ve watched this week.
The story comes from this AI creator who walked through what he found when he searched “ChatGPT” on the Apple App Store. The real ChatGPT app didn’t even appear in the top results. The whole first page was knockoffs with names like “GBT 5.4” and “Ask AI Anything,” all using logos that mimic OpenAI’s branding. Most likely they’re routing users to the free tier of ChatGPT and pocketing the subscription fee.
🛒 Why this matters
This isn’t a one-off. The creator pointed out that for almost any popular AI tool, scam clones flood the App Store. Less technical users like parents or grandparents have no way to tell the real app from the fake. Apple’s search ranking does nothing to protect them. And once someone pays, getting a refund is doable but annoying.
🔍 How to spot a fake AI app
The original poster’s quick checklist, based on what tripped his mom up:
- Check the developer name. Real ChatGPT is published by OpenAI. Real Claude is by Anthropic. Anything else is suspicious.
- Watch for vague phrases like “powered by GPT-5” or “built on GPT.” Legit apps don’t need to advertise the underlying model that way.
- Skip the App Store search. Go to the actual company website (openai.com, anthropic.com, perplexity.ai) and download from there.
- Look at the review count. Real apps have millions. Clones have a few thousand padded reviews.
💡 The bigger AI takeaway from the stream
The creator also broke down a recent Andrej Karpathy talk at Sequoia’s AI Ascent event about why AI feels “jagged”: brilliant at coding and math, oddly dumb at simple logic puzzles. Two reasons stuck with me:
- Verifiability. Code either runs or it doesn’t. That tight feedback loop makes reinforcement learning easy. Creative writing or common-sense reasoning has no clean signal.
- Revenue incentives. Labs train hardest on what pays, which is coding and math. Everything else gets less attention.
He demoed this live by asking GPT-5.3 “my car wash is 50 meters away, should I walk or drive?” The model told him to walk. Only after prompting it to act as a logical reasoning expert did it figure out the car needs to be at the car wash. Gemini 3 nailed it on the first try.
🛠️ Quick tips from the stream
- For older relatives, bookmark the real ChatGPT, Claude, and Perplexity websites directly on their home screen. Skip the App Store entirely.
- If a charge looks fraudulent, dispute it with your credit card. The host noted Apple Pay refunds are usually quick too.
- When a model gives you a dumb answer, try adding “act as a logical reasoning expert” before rephrasing. Cheap trick that often works on non-thinking models.
📺 Want the full breakdown?
The stream covered way more, including the Anthropic token controversy, Theo’s viral open letter, and Meta tracking employee keystrokes to train its models. Worth watching the full video for the deeper takes.